Non-contrast ultrasonic micro blood flow function imaging method and system

Abstract

The invention discloses a non-contrast ultrasonic micro blood flow function imaging method and system. The system comprises an ultrasonic signal acquisition module A1; the ultrasonic image processing module A2 is used for processing the collected ultrasonic image, filtering tissue signals and noise signals and enhancing blood flow signals; the ultrasonic blood flow signal extraction module A3 is used for performing micro blood vessel recognition and blood flow signal extraction in the enhanced ultrasonic blood flow image sequence; the micro blood vessel blood flow function quantitative evaluation module A4 is used for calculating blood flow function indexes of independent micro blood vessels according to the extracted micro blood vessel blood flow signals; the micro blood vessel function network construction module A5 is used for constructing a micro blood vessel function network according to the micro blood vessel blood flow signals and the corresponding blood flow function indexes; and the micro blood vessel function network index estimation module A6 is used for calculating micro blood vessel static network and dynamic network indexes. The system provides a new tool for brain function analysis by enhancing and accurately identifying tiny blood flow in an ultrasonic image sequence.

Description

Technical Field

[0001] The present invention relates to ultrasonic functional imaging and image post-processing and analysis technology thereof, and in particular to a method and system for non-contrast microvascular ultrasonic functional imaging and processing. Background Art

[0002] In the brain, to support the metabolic needs of billions of neurons, the cerebrovascular system, as a complex network, ensures fine regulation of blood supply through precise spatiotemporal regulation of cerebral blood flow (CBF), a phenomenon known as neurovascular coupling (NVC). Most neuroimaging methods utilize the neurovascular coupling mechanism to reflect the functional response of the brain by measuring local changes in blood flow or oxygenation during neural activity.

[0003] Functional ultrasound imaging (fUS), proposed by Emilie Macé et al. in 2011, is a typical example. The basic principle of this method is to obtain high-frame rate ultrasound signals through multi-angle coherent composite plane wave technology, and then high-pass filter the obtained ultrasound image to remove tissue signals after beam synthesis and IQ demodulation to obtain ultrasonic blood flow signals. This technology is also known as ultrafast Doppler technology. It has received widespread attention in recent years due to its higher temporal and spatial resolution than other brain functional imaging methods, but this method still cannot meet the needs of high-precision imaging of tiny blood flows. Physiological studies have shown that neurovascular coupling mainly occurs in small arteries with a scale of 20-200 microns, which are the basic units of NVC. Therefore, if you want to conduct in-depth research on brain function based on the neurovascular coupling mechanism, you must perform high-precision imaging of tiny blood flows in the brain.

[0004] In recent years, the ultrasound localization microscope (ULM), inspired by optical super-resolution imaging technology, overcomes the diffraction limit of traditional ultrasound imaging by accurately locating and tracking intravascular microbubbles. Through ultrasound super-resolution technology, people can obtain high spatial resolution images of microvascular structures. However, ULM imaging requires the accumulation of low-concentration microbubbles in the blood vessels for a long time, resulting in a long time to reconstruct a single-frame super-resolution image, which is not suitable for functional imaging. Recently, Noémi Renaudin et al. proposed the fULM technology. Based on the classic ULM technology, by repeatedly collecting ultrasound echoes under hundreds of the same tasks, super-resolution vascular reconstruction can be achieved in a shorter time. However, although the current fULM method can reflect blood flow information at the scale of microvessels, its shortcomings are that low-concentration sparse contrast is required for imaging, and it can only provide a time resolution of about 1 Hz. In particular, the signal obtained by multiple averaging cannot reflect the dynamic blood flow response under real tasks.

[0005] At present, ultrasound-based detection of microblood flow signals and high-resolution imaging mostly rely on contrast agents to enhance exogenous blood flow signals. However, ultrasound contrast imaging also has applicability issues in many clinical scenarios. On the other hand, contrast-based ULM and fULM ultrasound microblood flow imaging sacrifice temporal resolution while improving spatial resolution, and cannot meet the basic requirement of functional imaging with a temporal resolution of at least 10Hz. This results in current contrast-enhanced ultrasound microblood flow imaging being able to capture only a few static blood flow function indicators (such as vascular diameter, blood perfusion, etc.), and being unable to perform overall brain function dynamics analysis. Conventional non-contrast ultrasound functional imaging is limited by spatial resolution and sensitivity, and can only use brain regions as units when performing brain function network analysis, making it difficult to focus on the true unit of neurovascular coupling—a single microvessel.

[0006] In fact, most functional imaging needs to be performed without contrast. The low signal-to-noise ratio at this time makes it easy for micro-blood flow signals to be buried in noise signals. In fact, contrast-free micro-blood flow ultrasound functional imaging can be regarded as a weak video signal detection problem. In recent years, although deep learning-based video signal detection and enhancement methods have made remarkable progress, such methods are not suitable for the processing and exploration analysis of personalized ultrasound functional imaging results due to the lack of an interpretable "black box" mode. Therefore, it is urgent to introduce an interpretable weak video signal detection method for contrast-free ultrasound micro-blood flow signal detection and imaging. Summary of the invention

[0007] The technical problem to be solved by the present invention is to propose a non-contrast ultrasound micro blood flow functional imaging method and system in view of the above-mentioned deficiencies in the prior art, which is used to achieve high-precision micro blood flow functional imaging and analysis without contrast agents.

[0008] The technical solution proposed by the present invention to solve the above problems is to provide a non-contrast ultrasound micro-blood flow imaging method and system, which is used to collect high frame rate ultrasound echo signals and enhance and identify the micro-blood flow signals therein during ultrasound imaging, extract the blood flow signals of independent micro-vessels and analyze them, including the following modules:

[0009] A1. Ultrasonic signal acquisition module: Use a high-frequency ultrasonic probe with a center frequency of not less than 2MHz to transmit multi-angle plane wave ultrasound, and synthesize high-frame rate and high signal-to-noise ratio ultrasound images through coherent composite plane wave technology, with a frame rate of not less than 100Hz. Obtain the whole brain high-frame rate ultrasound echo signal J0.

[0010] A2. Ultrasonic image processing module: Based on the ultrasonic echo signal J0, the blood flow signal is enhanced by the micro-blood flow enhancement algorithm X1, and the background tissue signal and the noise signal are suppressed. The micro-blood flow enhancement algorithm is composed of a wall filtering algorithm X11 and a weak signal detection algorithm X12 based on a nonlinear dynamic model. In the present invention, the wall filtering algorithm is preferably SVD filtering, and the nonlinear dynamic model selected by the weak signal detection algorithm is preferably the Duffing double oscillator model. The micro-blood flow enhancement algorithm X1 includes the steps of:

[0011] (a) The ultrasonic echo signal J0 is transformed into a Casorati matrix S, and a singular value decomposition is performed on S to obtain a left singular matrix U, a right singular matrix V and a singular value matrix D;

[0012] (b) from the singular value matrix D, set several larger singular values ​​to 0 to obtain a new singular value matrix D';

[0013] (c) Perform inverse transformation on the new singular value matrix D' and the left singular matrix U and the right singular matrix V to obtain S';

[0014] (d) transforming S' into an ultrasonic blood flow signal sequence J0';

[0015] (e) Extract the time series curve j0(t) at each pixel position from the ultrasonic blood flow signal sequence J0', and perform unbiased processing on it to obtain j0(t)';

[0016] (f) j0(t)' is used as the input signal of the Duffing double-oscillator model H, and the output desynchronized time series is used as the enhanced ultrasonic blood flow signal j1(t).

[0017] In the above micro blood flow enhancement algorithm, the system expression of the Duffing double oscillator model H is:

[0018]

[0019] Where F is the amplitude of the periodic driving force, ω is the angular frequency of the periodic driving force, ξ is the damping coefficient, and k is the coupling coefficient.

[0020] Thus, the enhancement processing of the ultrasonic micro blood flow signal is completed.

[0021] A3. Ultrasonic blood flow signal extraction module: Based on the enhanced ultrasonic blood flow image sequence J1, each single blood vessel in the ultrasonic blood flow image is accurately identified and the blood flow signal is extracted through the pattern recognition algorithm X2, including the steps of:

[0022] (a) The enhanced ultrasonic blood flow image sequence J1 is accumulated and averaged frame by frame to obtain a high signal-to-noise ratio ultrasonic blood flow image O;

[0023] (b) The high signal-to-noise ratio ultrasound blood flow image O is binarized by an adaptive threshold binarization algorithm to obtain the roughly identified initial candidate vascular region Zq;

[0024] (c) Select any pixel point q that has not been selected in Zq, calculate the correlation measure between the time series corresponding to point q and the time series corresponding to the neighboring pixels of q in the average graph O, and update Zq accordingly, that is, mark the pixel points greater than the threshold as candidate blood vessel regions Zq;

[0025] (d) Repeat step (c) and continuously update Zq until all pixel positions in Zq are traversed;

[0026] (e) treating all connected candidate vascular regions as an independent vascular region and marking all independent vascular regions one by one;

[0027] (f) Screening all the marked independent blood vessel regions, retaining the blood vessels with an average diameter below the set threshold as microvascular regions, and obtaining a structural map J2 containing K independent microvessels;

[0028] (g) Based on J2, the blood flow signals corresponding to each pixel in the k-th single vessel region are superimposed and averaged to obtain a single time series, which is the blood flow signal S corresponding to the single vessel. k (k=1,…,K).

[0029] In the above single blood vessel recognition algorithm, the adaptive binarization algorithm is preferably the Otsu method.

[0030] In the above single vessel identification algorithm, the correlation metric is preferably:

[0031]

[0032] Where xcorr(*) represents the cross-correlation of two time series, x and y represent the time series signals corresponding to two pixels, τ represents the lag coefficient, and (-τ0, τ0) is the preset range of lag coefficient values.

[0033] Therefore, by comparing the temporal and spatial correlations between pixels, accurate identification of single blood vessel regions is achieved, and a structural image J2 that clearly distinguishes K independent microvessels is obtained.

[0034] A4. Microvascular blood flow signal quantitative evaluation module: based on the independent microvascular blood flow signal S k , calculate the independent functional index of independent microvessels. Specifically, S can be calculated by the classic time series feature algorithm. k The time domain characteristics or frequency domain characteristics of the microvascular blood flow function are used as independent microvascular blood flow function indicators. These function indicators can be perfusion index, response delay, pulsatility index, power spectral density, information entropy, etc.

[0035] A5. Microvascular function network construction module: Based on the extracted blood flow signals and the calculated microvascular function indicators, a microvascular function network is constructed. Specifically, each independent microvascular is used as a node of the network, and the microvascular signal S k Or based on S k The calculated blood flow function index is used as the attribute of the corresponding node, and the connectivity between nodes is determined based on the statistical dependence of blood flow signals or functional characteristics between all K independent microvessels, thereby constructing a microvascular functional network N. The statistical dependence used to determine the connectivity between nodes can be classical correlation, mutual information, transfer entropy, coherence, Granger causality, and the coherence recurrence graph index and the average length of adjacent tracks of the double chaotic time series in the phase space.

[0036] A6. Microvascular functional network index calculation module: Based on the constructed microvascular functional network N, according to the classic static network and dynamic network index estimation method, the corresponding static index or dynamic evolution index of the functional network N is calculated. The microvascular network function index can be connection strength, clustering coefficient, and number of strongly connected components.

[0037] The above-mentioned technical solutions are adopted to facilitate high-precision blood flow signal recognition and extraction without contrast imaging, quantitative evaluation and functional analysis of tiny blood flow signals, and improve the accuracy and reliability of ultrasonic functional imaging and analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 This is a schematic block diagram of the overall process of a non-contrast ultrasound micro-blood flow functional imaging system according to an embodiment of the present invention;

[0039] Figure 2 is a schematic block diagram of a processing algorithm for enhancing a micro blood flow signal according to an embodiment of the present invention;

[0040] Figure 3 This is a schematic block diagram of an ultrasonic micro-blood flow signal extraction module processing according to an embodiment of the present invention;

[0041] Figure 4 It is a schematic diagram of the single-vessel blood flow signal recognition and extraction effect according to an embodiment of the present invention. DETAILED DESCRIPTION

[0042] The present invention is further described in detail below in conjunction with the accompanying drawings:

[0043] Figure 1 is an overall flow chart of the non-contrast ultrasound micro-blood flow imaging method of the present invention, such as Figure 1 As shown:

[0044] In module A1, high frame rate ultrasonic signals are collected by a high frequency ultrasonic transducer.

[0045] In a specific embodiment of the present invention, an ultrasonic probe with a center frequency of 30 MHz is optionally used to transmit a 9-angle plane wave, and the ultrasonic plane wave transmission frame rate is 4500 Hz. After coherent compounding, an ultrasonic echo signal with a frame rate of 500 Hz is obtained. The 500 Hz frame rate meets the time resolution requirement for subsequent dynamic analysis of blood flow signals.

[0046] In module A2, the collected ultrasound signal is processed by a micro-blood flow enhancement algorithm to significantly enhance the blood flow signal component.

[0047] The flow chart of the micro-blood flow enhancement algorithm is as follows Figure 2 As shown:

[0048] First, the ultrasound IQ signal matrix is ​​transformed into the Casorati matrix form S. One column in the Casorati matrix corresponds to one frame of IQ signal. SVD filtering operation is performed based on S. The tissue signal and blood flow signal in the ultrasound signal can be separated by SVD due to their different temporal and spatial characteristics. The tissue signal has stronger energy and corresponds to several larger components after SVD. Filtering it can effectively suppress the tissue signal and highlight the blood flow signal.

[0049] In a specific embodiment of the present invention, the filter coefficient of the SVD filter is set to 5%, that is, the first 5% of the spatiotemporal components of the SVD are filtered out.

[0050] After transforming the filtered matrix S' into an IQ signal form, the complex signal is modulo-calculated to obtain an ultrasonic blood flow image sequence, and the time series signal at each pixel position in the image sequence is unbiased, that is, the initial value of the signal is subtracted, and then it is input into the Duffing double oscillator model for denoising. This is because the phase space trajectory of the Duffing oscillator is sensitive to weak signals but not affected by noise. When the weak signal time series is used as the input of one of the oscillators of the Duffing double oscillator model, it will be out of synchronization with the other oscillator. The out-of-synchronization time series is the enhanced weak signal. Therefore, with the help of the Duffing double oscillator model, the noise signal can be filtered out while detecting the micro blood flow signal.

[0051] In a specific embodiment of the present invention, the parameters of the Duffing double oscillator model are selected as follows: the periodic driving force amplitude F is 1, the periodic driving force circular frequency ω is 2, the damping coefficient ξ is 0.8, and the coupling coefficient k is 10.

[0052] In module A3, a single blood vessel recognition algorithm is used to perform high-precision recognition of a single micro-vessel in an ultrasonic blood flow image sequence, and on this basis, a single blood vessel blood flow signal extraction algorithm is used to extract the blood flow signal of each micro-vessel.

[0053] The processing flow chart of the ultrasonic blood flow signal extraction module is as follows: Figure 3 As shown:

[0054] The ultrasonic blood flow image sequence is cumulatively evaluated frame by frame to obtain a blood flow image O with a high signal-to-noise ratio, and O is binarized. Since the blood flow image has been previously subjected to tissue signal filtering, noise filtering, and blood flow signal enhancement, the part with a higher pixel grayscale value in the image should theoretically be the area where the blood flow is located, while the grayscale value of the remaining area should be lower. The area where the blood vessel is located can be preliminarily segmented by binarization. In order to further perform more precise single-vessel recognition, the present invention introduces a spatiotemporal correlation prior, assuming that the signal in the area where the independent blood vessel is located should have a higher spatiotemporal correlation, and optimizes the independent blood vessel area by measuring the correlation of the corresponding time series signals between pixels. Since the present invention mainly focuses on the blood flow signals of microvessels, after obtaining a series of independent blood vessel areas, large blood vessels are eliminated by screening the average diameter of the independent blood vessels.

[0055] In a specific embodiment of the present invention, a pixel position is selected in the initially segmented candidate independent vascular region Zq and its correlation measure with the time series signal corresponding to the adjacent pixel position is compared. The pixel position greater than the threshold is selected and updated to Zq and all the pixel positions in Zq are continuously traversed to achieve fine recognition of the independent vascular region. Specifically, the correlation measure is selected

[0056]

[0057] That is, the maximum cross-correlation index of the time series signal corresponding to adjacent pixels within a certain delay range, the maximum delay τ0 is selected as 10ms, corresponding to 500Hz or 5 frames; the correlation threshold is selected as 0.6. The finely identified independent vascular regions are further screened, and the independent vascular regions with an average diameter of less than 200 microns are retained as microvessels, and finally a structural map J2 containing K independent microvessels is obtained.

[0058] Based on the K identified independent microvascular regions, the time series signals corresponding to the pixels in each region are averaged to extract the blood flow signal S of the independent microvascular. k (k=1,…,K).

[0059] Figure 4 The figure above is a schematic diagram of the single blood vessel identification and micro blood flow signal extraction effect obtained by the above method of the present invention. The upper left figure is the ultrasound image enhanced by the above module A2, and the lower left figure is the micro blood vessel area identified after processing by the above module A3. Based on this, the blood flow signals of the four representative independent micro blood vessels in the figure are extracted, as shown in the figure below. Figure 4 As shown in the middle right side, the blood flow signals of the four blood vessels can effectively reflect the heart rhythm in the resting state.

[0060] Then, based on the extracted Figure 4The K single-vessel blood flow signals S shown on the right k (k=1,…,K), and brain function indicators are calculated through module A4. Since brain function is reflected in the blood perfusion of a single blood vessel on the one hand, and the coupling of blood flow activities between different blood vessels on the other hand, the independent function indicators of a single blood vessel and the network function indicators of the microvascular network are calculated respectively, so as to achieve a comprehensive and reliable brain function analysis.

[0061] In a specific embodiment of the present invention, the independent functional index of a single blood vessel is calculated as follows: the average amplitude of the blood flow signal of a single blood vessel is calculated as the blood perfusion coefficient, and the correlation between the blood flow signal of a single blood vessel and the stimulus event signal is calculated as the functional response coefficient; the microvascular network is constructed as follows: a single blood vessel is used as a node of the network, and the correlation of the blood flow signal corresponding to the single blood vessel is used as the basis for measuring the connectivity between nodes, that is, connectivity is established for microvessels with high correlation, and connectivity between nodes with low correlation is ignored, thereby obtaining a microvascular network connectivity matrix, and the number of strongly connected components and the clustering coefficient are calculated based on the matrix as microvascular network functional indicators.

Claims

1. A method and system for non-contrast ultrasound micro-blood flow functional imaging, characterized in that: For ultrafast ultrasound video images without contrast agents, a nonlinear dynamic model is used to detect weak signals, and then the spatiotemporal characteristics are used to extract independent microvessels. Each independent microvessel is used as a unit for quantitative hemodynamic evaluation and network function evaluation. The system consists of the following modules: (a) Ultrasonic signal acquisition module A1: uses ultrafast ultrasonic imaging technology to acquire the whole-brain high-frame-rate ultrasonic echo signal J0; the probe center frequency is not less than 2MHz, and the frame rate is not less than 100Hz; (b) Ultrasonic image processing module A2: for J0, the background tissue signal and noise signal are suppressed by the commonly used wall filtering algorithm X0; then the blood flow signal is contrast enhanced by the weak signal detection method X1 based on the nonlinear dynamic model to obtain the contrast enhanced ultrasonic blood flow image sequence J1; (c) Ultrasonic blood flow signal extraction module A3: Based on J1, through the pattern recognition algorithm X2, a structural diagram J2 that clearly distinguishes K independent microvessels and the corresponding K independent microvessel blood flow signals S are obtained. k (k=1,…,K); (d) Microvascular blood flow function quantitative evaluation module A4: based on S k , according to the single time series time domain and frequency domain feature algorithms, the blood flow function index of each independent microvessel is calculated; (e) Microvascular functional network construction module A5: Each independent microvascular is regarded as a node of the network. k Or based on S k The calculated blood flow function index is used as the attribute of the corresponding node; the connectivity between nodes is determined based on the statistical dependence of blood flow signals or functional characteristics between all K independent microvessels, and a microvessel functional network N is constructed; (f) Microvascular functional network index estimation module A6: According to the classical static network and dynamic network index estimation methods, the corresponding static index or dynamic evolution index of the functional network N is obtained.

2. The weak signal detection algorithm X1 as claimed in claim 1, characterized in that: The Duffing double oscillator model is the preferred nonlinear dynamic model: Wherein, F is the amplitude of the periodic driving force, ω is the angular frequency of the periodic driving force, ξ is the damping coefficient, and k is the coupling coefficient; the input of the Duffing double-oscillator model is the time series signal at each pixel position; the output of the Duffing double-oscillator model is the double-oscillator desynchronization time series, and the desynchronization time series of all pixel positions constitute the contrast-enhanced ultrasonic blood flow image sequence J1.

3. The pattern recognition algorithm X2 as claimed in claim 1, characterized in that: By using the spatiotemporal similarity inside independent blood vessels, the structural graph J2 containing K independent microvessels and the corresponding K independent microvessel blood flow signals S are automatically obtained. k (k=1,…,K); the specific steps are: (1) The enhanced ultrasonic blood flow image sequence J1 is superimposed and averaged frame by frame to obtain a high signal-to-noise ratio average ultrasonic blood flow image O; (2) Adaptively binarize the average ultrasonic blood flow image O to obtain a roughly identified foreground vessel map, which is used as the initial candidate vessel region Zq; (3) Automatically select an unselected pixel point q in Zq, use the constructed correlation metric to test the spatiotemporal similarity between the position of point q and the neighboring pixel positions of q in the average graph O, and update Zq. That is, select the neighboring pixels whose spatiotemporal similarity is greater than the preset threshold and mark them as candidate vascular regions Zq. (4) Repeat step 3) in the updated Zq until all pixels in Zq are traversed; (5) Treat all connected candidate vascular regions as an independent vascular region, and obtain all independent vascular regions one by one; (6) Screening blood vessels with an average diameter below a set threshold as microvessels, and obtaining a structural graph J2 containing K independent microvessels; (7) The average signal of all pixels in each independent microvascular region in J2 is represented as the blood flow signal of the current microvascular; finally, the blood flow signals S of K single blood vessels are obtained. k (k = 1, 2, …, K).

Images